binding

binding ¶

IVIM binding adapter for the shared fitting infrastructure.

BoundIVIMModel wraps an IVIMModel together with b-values, producing a FittableModel that the shared fitter can use without knowing about IVIM-specific context.

Provides an analytical Jacobian for efficient Levenberg-Marquardt optimization.

BoundIVIMModel ¶

BoundIVIMModel(
    model, b_values, fixed=None, b_threshold=200.0
)

Bases: BaseBoundModel

IVIM model with b-values bound.

Wraps an IVIMModel so the fitter only sees predict_array_batch(free_params) -> output.

PARAMETER	DESCRIPTION
`model`	IVIM signal model. TYPE: `IVIMModel`
`b_values`	Diffusion weighting values in s/mm^2. TYPE: `NDArray`
`fixed`	Parameters to fix at constant values during fitting. TYPE: `dict[str, float] \| None` DEFAULT: `None`
`b_threshold`	b-value threshold for segmented initial guess estimation. TYPE: `float` DEFAULT: `200.0`

ensure_device ¶

ensure_device(xp)

Transfer b-values array to the target device.

predict_array_batch ¶

predict_array_batch(free_params_batch, xp)

Predict signal for a batch of voxels.

PARAMETER	DESCRIPTION
`free_params_batch`	Free parameter values, shape `(n_free, n_voxels)`. TYPE: `NDArray`
`xp`	Array module. TYPE: `module`

RETURNS	DESCRIPTION
`NDArray`	Predicted signal, shape `(n_b, n_voxels)`.

get_initial_guess_batch ¶

get_initial_guess_batch(observed_batch, xp)

Get initial parameter guesses using segmented approach.

Uses log-linear fit at high b-values for D, heuristic for f and D*, and b=0 signal for S0.

PARAMETER	DESCRIPTION
`observed_batch`	Observed data, shape `(n_b, n_voxels)`. TYPE: `NDArray`
`xp`	Array module. TYPE: `module`

RETURNS	DESCRIPTION
`NDArray`	Initial guesses, shape `(n_free, n_voxels)`.

compute_jacobian_batch ¶

compute_jacobian_batch(params_batch, predicted, xp)

Compute analytical Jacobian for IVIM bi-exponential model.

Only computes columns for free (non-fixed) parameters.

PARAMETER	DESCRIPTION
`params_batch`	Free parameter values, shape `(n_free, n_voxels)`. TYPE: `NDArray`
`predicted`	Predicted signal, shape `(n_b, n_voxels)`. TYPE: `NDArray`
`xp`	Array module. TYPE: `module`

RETURNS	DESCRIPTION
`NDArray`	Jacobian, shape `(n_free, n_b, n_voxels)`.